Shopify CEO Tobi Lütke is threatening to prohibit Claude Code at his company unless Anthropic changes how the AI coding agent reads configuration files, according to posts he made on X this week. The issue centers on a Markdown file format disagreement that, while seemingly minor, creates what Lütke calls a "split brain problem" when thousands of developers work in Shopify's massive shared codebase. Claude Code relies on CLAUDE.md files for project-specific instructions, but refuses to natively support the industry-standard AGENTS.md format that most other coding tools have adopted.
The technical divergence creates real operational headaches in large repositories. Shopify employs thousands of engineers working in a single monorepo, and because they don't all use identical AI coding assistants, each agent needs to locate the correct instructions for whichever section of the codebase it's currently handling. When one coding agent ignores the same instruction files that others read, developers end up with agents running under conflicting rules. Anthropic offers workarounds including importing AGENTS.md from a CLAUDE.md file or using symbolic links, but in a large repository, teams still must ensure those instructions remain synchronized across all locations where they appear. A 2026 study examining 2,926 GitHub repositories discovered that context files have become the most prevalent method developers use to give coding agents instructions, with AGENTS.md already deployed across multiple different tools.
"I'm thinking about banning Claude Code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.," Lütke posted on X. He explained that "agents and Claude files are recursively applied through the tree," meaning instruction files can exist throughout the directory structure to provide an agent with different guidance depending on what it's working on. When a directory contains instructions for tools reading AGENTS.md but lacks equivalent instructions for Claude Code, developers using Claude can find themselves operating without the same context — what Lütke described as working "with lobotomy." The report notes that Shopify has already constructed automation to address the discrepancy, but Lütke stresses that "it's a stupid complexity tax that shouldn't have to be paid."
OpenAI introduced AGENTS.md in August 2025 to provide coding agents with project-specific instructions, and by December it was being used by more than 60,000 open-source projects and agent frameworks, with backing from tools including Codex, Cursor, Gemini CLI, GitHub Copilot, Jules, and VS Code. OpenAI later transferred the format to the Agentic AI Foundation under the Linux Foundation. Developers have been requesting that Anthropic support AGENTS.md for nearly a year, with more recent requests continuing to push for native support, including one focused on recursive AGENTS.md discovery, which Anthropic has closed as "not planned." As companies bring more coding agents into the same repositories, keeping those tools aligned becomes one more responsibility for platform teams — a burden that grows heavier when vendors insist on proprietary configuration approaches despite industry convergence around a common standard. The CEO's public ultimatum signals that enterprise customers may increasingly treat interoperability not as a nice-to-have feature but as a prerequisite for adoption, forcing AI companies to choose between maintaining control over their ecosystems and securing access to the largest development organizations.